Codesota · Benchmark · LVIS v1.0Home/Leaderboards/Vision & Documents/Object Detection/LVIS v1.0
Unknown

LVIS v1.0.

1,203 object categories with federated, long-tail distribution across 164K COCO images. Tests real-world detection with rare and fine-grained categories.

Paper ↗Leaderboard ↓
§ 01 · SOTA history

Year over year.

§ 02 · Leaderboard

Results by metric.

Found a wrong score or missing run?
Use row edits to send a sourced correction into moderation.
Add / edit result ↗Report issue ↗

Box Ap

Box Ap is the reported evaluation metric for LVIS v1.0. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for Box Apverifiedpapervendorcommunityunverified
RankModelTrustScoreYearLinksFix
01DINO-X
DINO-X box AP on LVIS v1.0 minival. Table 3, arxiv:2411.14347.
verified71.42024Paper ↗Looks wrong?
02APE-Large
APE-Large box AP on LVIS v1.0 minival. Table 2, arxiv:2312.02153.
verified70.32023Paper ↗Looks wrong?
03Co-DINO (ViT-L)
Co-DINO with ViT-L backbone (Objects365 pre-training). 68.0 box AP on LVIS v1.0 val. ICCV 2023. Also reported: 67.9 box AP on LVIS val in the abstract. arxiv:2211.12860.
verified682022Paper ↗Looks wrong?
04ViTDet-H (MAE)
ViTDet-H box AP on LVIS v1.0 minival. Table 3, arxiv:2203.16527.
verified642022Paper ↗Looks wrong?

Mask Ap

Mask Ap is the reported evaluation metric for LVIS v1.0. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for Mask Apverifiedpapervendorcommunityunverified
RankModelTrustScoreYearLinksFix
01DINO-X
DINO-X unified model. 67.0 mask AP on LVIS v1.0 minival — SOTA at time of release Nov 2024. Table 3, arxiv:2411.14347.
verified672024Paper ↗Looks wrong?
02APE-Large
APE-Large with EVA-02 ViT-L backbone. 66.4 mask AP on LVIS v1.0 minival in closed-set evaluation. Table 2, arxiv:2312.02153. CVPR 2024.
verified66.42023Paper ↗Looks wrong?
03InternImage-H
InternImage-H (1B params) with DINO detection head. 65.4 mask AP on LVIS v1.0 minival. Table 5, arxiv:2211.05778. CVPR 2023.
verified65.42022Paper ↗Looks wrong?
04EVA-02-L
EVA-02-L with Cascade Mask RCNN. 62.1 mask AP on LVIS v1.0 minival. Table 5, arxiv:2303.11331. CVPR 2023.
verified62.12023Paper ↗Looks wrong?
05ViTDet-H (MAE)
ViTDet-H with MAE pretraining and Cascade Mask RCNN head. 59.5 mask AP on LVIS v1.0 minival. Table 3, arxiv:2203.16527. ECCV 2022.
verified59.52022Paper ↗Looks wrong?
06Mask2Former (Swin-L)
Mask2Former with Swin-L backbone. 56.1 mask AP on LVIS v1.0 minival. Table 7, arxiv:2112.01527. CVPR 2022.
verified56.12021Paper ↗Looks wrong?
07ViTDet-H
ViTDet-H Cascade Mask R-CNN. LVIS v1.0 val. NeurIPS 2022.
paper53.42026Source ↗Looks wrong?
08EVA-02-L (LVIS)
EVA-02-L. LVIS v1.0 val. Strong long-tail performance.
paper50.72026Source ↗Looks wrong?
09Mask2Former (Swin-L) LVIS
Mask2Former Swin-L. LVIS v1.0 instance segmentation.
paper44.62026Source ↗Looks wrong?

Mask Ap Rare

Mask Ap Rare is the reported evaluation metric for LVIS v1.0. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for Mask Ap Rareverifiedpapervendorcommunityunverified
RankModelTrustScoreYearLinksFix
01APE-Large
APE-Large AP_rare on LVIS v1.0 minival. Table 2, arxiv:2312.02153.
verified65.42023Paper ↗Looks wrong?
02EVA-02-L
EVA-02-L AP_rare on LVIS v1.0 minival — strong rare category improvement. Table 5, arxiv:2303.11331.
verified612023Paper ↗Looks wrong?
03Mask2Former (Swin-L)
Mask2Former Swin-L — AP_rare on LVIS v1.0 minival. Table 7, arxiv:2112.01527.
verified53.52021Paper ↗Looks wrong?
§ 04 · Submit a result

Add to the leaderboard.

← Back to Object Detection